Don’t take anything for granted man, life is too unpredictable.
Death is really strange, because the loss itself isn’t what hurts. Rather, the regret of unmet plans and the pain of missing out on future conversations.
Thinking about time we should’ve spent together and won’t.
@polsia is literally AI Slop (seriously, read it backwards) Was that how it was named? I’ve never seen spam at this level. Some VCs are really throwing their money at the AI equivalent of a 2000s era programmatic SEO spammer. Are there any examples where scam startups won?
@RhysSullivan@colemurray OpenInspect + Sentry log listening + risk profile review to determine merge safety makes this pretty straightforward already. We aren’t quite there yet as an org but I’m sure some are
@nbaschez We do this for technical assessments and it’s really just a litness test for familiarity with modern tools.
We proctor a brief writing portion (design doc), which really separates candidates. They’re allowed to use agents for that too, but we monitor how they’re used.
We’ve automated every single thing we can @every with AI agents.
And yet there’s way more human work to do than ever. We’ve gone from 4 -> 30 human employees since GPT-3.
I wrote a report on the structural reasons: how AI makes expert competence cheap, why that drives up demand for experts, and why the dynamic only intensifies as we approach AGI.
After Automation: https://t.co/Lb7SUCduAg
I’m convinced most of these crazy long /goal runs are just poorly optimized environment instructions. So easy for the model to work for 12+ hrs if the tool calls its waiting on are CI checks (which you should probably move to be local) Most I’ve seen is 6hrs w/ a large scope goal
Fork your dependencies, trim them to only your use case, never update unless it breaks for your users. I’ve been vocal about this for 10+ years. I’ve always said that updating is way riskier than latent bugs (which can be tracked and CVEs monitored).
If you are updating a dependency, it’s on you to analyze every single commit in the full transitive set of dependencies. If you dont see anything compelling, dont update!
I remember at HashiCorp once in awhile an engineer would try to update a dep or replace a DIY lib with an external one and id always ask “show me the commit we need.” Dont update for the sake of it.
Feeling pretty swell about this mentality with all the supply chain attacks happening.
Seriously, when are we going to stop with these dynamic reasoning effort / auto routing attempts? They don’t work and when I ask the model to do something I want the model to try as hard as it possibly can, at least when I ask! I wish I could explicitly set the reasoning effort like I can in Claude code.
Claude mobile with Opus 4.7 & adaptive thinking has become unusable. Refuses to think or call tools, even when I’ve explicitly asked for web searches. I can barely get tool calls to happen anymore. I am dumbfounded by the outright refusal. $200/mo ?! Wdym you won’t try?!
At that age, you’re so plastic, just read a lot and always study how other projects have done something! If you’re working independently in a fan of taking on something really big and trying to build something useful. I built a few websites in high school for friends & clubs and I couldn’t stop working on them or thinking about it. Don’t overcomplicate it imho; Matt’s skills have done well because they leverage decades of foundational SWE wisdom - using these skills as she’s learning to build will work. I believe there’s a lot of value to reading; get her a copy of Code Complete 2nd edition (GOAT programming book) and it’ll become clear what’s important to pay attention to while building. All leaders are readers!
This is just one of my favorite topics. It’s obviously a good time to be starting your career in SWE, but most of the discourse is negative!
As with anything - you get out what you put in! If you’re not exhausted by the end of writing the design doc when you’re a junior, you’re doing it wrong. It’s kind of like building your vocabulary; when you come across a word you don’t understand you should look it up & retain its definitions before moving on. Sure, you might pick it up from context clues, but that’s really not good enough for pursuing mastery.
Agents are still pretty bad for non-coding use cases. I ran a workflow for an upcoming trip to London, selecting outfits based on my itinerary. I was really hoping that with Codex + Chrome plugin it could search, evaluate, and add to cart. Only kind of works, very lazy selections